The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and process schema metadata and GraphQL introspection data from external endpoints to generate application code and queries.
Ingestion points: In references/query-runtime.md, the skill describes fetching _meta contract data and standard GraphQL introspection results from data endpoints.
Boundary markers: The instructions mention validation via assertMetaContract and assertMetaQuery but do not specify explicit delimiters or "ignore embedded instructions" warnings for the data being interpolated into code generation prompts.
Capability inventory: The skill environment includes network access (via fetch) and file generation capabilities (via the referenced graphql-codegen package).
Sanitization: Data integrity is checked using contract assertions, but there is no explicit mention of sanitizing metadata values before they are used in code templates.
[DYNAMIC_EXECUTION]: The skill documentation describes a utility for dynamically generating and loading code during testing.
Evidence: references/codegen-orm-output.md highlights the runCodegenAndLoad helper from the @constructive-io/graphql-test package, which is designed to "generate ORM code and dynamically load it in a single call" for use within test suites.

constructive-orm